Episode 37: AI as the Learned Colleague, Not the Ghostwriter

What does it look like to write an academic paper with AI without letting AI write a single word of it? Craig and Rob open with that question, then move through a new UK survey on how student AI use is evolving, ChatGPT's shift from Pulse to Scheduled Tasks, and fresh labor market data on what AI is doing to entry-level jobs.

Craig describes using Codex to co-produce a 25-page conference paper in about three days, not by asking the tool to write sections, but by writing them himself and requesting feedback, the same back-and-forth he'd have with a co-author. That framing anchors a wider conversation about inconsistent AI disclosure rules across journals and conferences, then the hosts turn to the Higher Education Policy Institute's 2026 GenAI survey of UK students, which shows near-universal AI adoption alongside a narrowing of specific use cases. They close with ChatGPT's new Scheduled Tasks feature, a PwC report on entry-level jobs requiring senior-level skills, and updates to NotebookLM's video and slide tools.

What you'll hear

Writing as a colleague, not a ghostwriter. Craig used Codex to draft a 25-page conference paper by writing sections himself and requesting feedback, rather than asking the AI to generate text, plus using it to locate and format references and diagnose a thin section of the argument.
Inconsistent AI policies across academic venues. Rob raises the problem of conferences with looser AI disclosure norms and journals with stricter ones, including an editor's account of a publisher flagging words like "consequentially" as a telltale sign of AI use.
The HEPI 2026 GenAI survey results. Any AI use among UK students rose from 66% to 92% year over year, while text generation specifically dropped from 64% to 56%. Barriers to use are falling too: fear of being accused of cheating dropped from 53% to 42%, fear of hallucinations from 51% to 35%, and institutional bans as a deterrent from 31% to 21%.
Least-skilled, most-confident. A case study in the HEPI report found that medical students working with AI-tuned virtual patient cases who were least skilled were also most confident in their AI use, prompting a discussion of algorithmic trust and critical thinking.
The loneliness question. Of surveyed students, 59% said generative AI has no impact on their loneliness, with the rest split almost evenly between feeling less lonely and more lonely. Rob raises concerns about long-term effects, drawing a parallel to how long it took research and regulation to catch up with social media's impact on youth.
ChatGPT Scheduled Tasks and the PwC jobs data. The hosts compare ChatGPT's Scheduled Tasks (successor to ChatGPT Pulse) to Claude Code routines and Codex automations, then turn to PwC's 2026 Global AI Jobs Barometer, which found AI-exposed entry-level jobs are seven times more likely to require senior-level skills like judgment, leadership, creativity, and face-to-face interaction.

Episode highlights

(03:34) Craig on model convergence: "The top model of yesterday is now the mediocre model of today."
(09:40) Craig on his conference paper workflow: "I want to really emphasize that AI did not write anything... it was really just having that colleague there."
(19:34) Craig on falling barriers to AI use in the HEPI data: fear of being accused of cheating "dropped to 42% from 53%," and fear of "getting false results and hallucinations dropped from 51% to 35%."
(20:37) Craig on AI detectors: "So stop. If you're doing that, stop. They don't work. They're bad."
(34:53) Craig reading PwC's finding: "AI exposed entry level jobs are now seven times more likely to require traditional senior skills such as judgment, leadership, creativity and face to face interaction."
(38:32) Rob on tacit knowledge: the real concern is "how do you learn the norms of an institution and how they operate and how you get things done," something he doesn't think AI can teach at the individual level.

References mentioned

Higher Education Policy Institute (HEPI), 2026 GenAI student survey, UK, more than 1,000 students surveyed (freely available report)
PwC, 2026 Global AI Jobs Barometer, based on more than 1 billion job ads globally and 2.4 million US entry-level roles
An unnamed arXiv preprint from a few months prior, arguing official labor statistics undercount AI disruption because the real impact falls on tacit knowledge transmission through early-career work
Grok (xAI), used by Rob to draft a family lease agreement
Codex (OpenAI), used by Craig as a writing collaborator for a conference paper
Microsoft Copilot (M365), used by Rob for document review and editing feedback
ChatGPT Scheduled Tasks and its predecessor, ChatGPT Pulse (OpenAI)
Claude Code routines and Codex automations, comparison points for scheduled AI briefings
NotebookLM (Google), including its short-video and slide-editing features
AI Resilient Learning Activities Database, an upcoming free repository from AI Goes to College, funded by a Just Business grant through the Louisiana Tech College of Business

Main takeaways

AI as colleague, not author: Craig used Codex to draft a conference paper by writing sections himself and asking for feedback, not by having it generate text
Academic AI disclosure rules vary sharply by venue, creating real friction for multi-venue submissions
UK student AI use hit 92%, even as basic text generation use dropped, pointing to more varied use cases
Fear-based barriers (cheating accusations, hallucinations, institutional bans) are all declining as students gain experience
Least-skilled medical students showed the most confidence in AI-tuned tools, a warning sign for critical thinking
Student-reported loneliness effects from AI split almost evenly between better, worse, and no change
PwC data shows entry-level jobs now require senior-level skills like judgment and creativity seven times more often
The apprenticeship pipeline that once taught tacit workplace knowledge may not survive as that entry-level work disappears

Links

Book of Common Prayer: https://www.episcopalchurch.org/wp-content/uploads/2021/02/book-of-common-prayer-2006.pdf
HEPI report: https://www.hepi.ac.uk/reports/student-generative-ai-survey-2026/

---

AI Goes to College is a podcast for higher education professionals trying to make sense of artificial intelligence in their classrooms, their research, and their institutions. Co-hosted by Craig Van Slyke and Rob Crossler, the show focuses on practical, evidence-based perspectives on AI in higher education without the hype.

For all things AI Goes to College, including the podcast, go to https://www.aigoestocollege.com/.

Mentioned in this episode:

AI Goes to College Newsletter

What students actually do with AI

AI Goes to College

What students actually do with AI